Remaining life prediction of turbofan engines based on RF-SA-SDCNN
In order to solve the problem of low accuracy of turbofan prediction at present,a method of turbofan residual life prediction based on RF-SA-SDCNN fusion was proposed.First,the multi-sensor long sequence data were smoothed exponentially and normalized to reduce errors due to different dimensions,ranges of values,and noise fluctuations,and the importance features of the multi-sensor signals were extracted using random forest algorithm.Then,a prediction model based on random forest algorithm and self attention mechanism combined with stacked dilation convolution neural network was constructed.Self attention mechanism enhanced contribution degree by assigning different weights to features,and stacked dilation convolution extracted time series features for regression analysis by enlarging the model's sensory field.The model prediction accuracy was improved by using the grid search optimization algorithm and the StratifiedKFold cross validation method.Finally,the CMAPSS data set was used to verify the effectiveness of the proposed method.The results show that the proposed method can effectively improve the accuracy of turbofan residual life prediction.
random forest algorithmself-attentive mechanismstacked neural networkGridSearchK-Fold cross-validationexponential smoothing
NETL
NSTL
万方数据
